01 · in pieces
Work arrives in pieces.
A request in a group chat. The ticket in another project. A file someone sent last week.
Message the task. Get back “Done.”
Lamp is an AI assistant for your team’s chats. Ask it in Telegram, WhatsApp or Slack. It updates Jira, keeps private notes for each person, and tells you what it did in a sentence.
youmove the launch ticket to in progress
lampDone. CLM-42 is in progress.
02 · lining up
Lamp lines them up.
It works out where the request belongs, finds the ticket, reads it before changing anything and checks the result after.
03 · one line back
You get one line back.
“Done. CLM-42 is in progress.”
Seen from the lamp, the scattered pieces make its mark.
where it lives
In the chat apps your team already uses.
Message Lamp in Telegram, WhatsApp or Slack, the way you’d message a colleague. There’s no new app to learn and no dashboard to keep open.
Telegramdirect message what's left on the launch?
Two tickets: CLM-44 and CLM-47.
WhatsAppchat send me the standup notes
Here they are.
standup-notes.md
Slack#product @Lamp who owns the login bug?
Priya Shah. MOB-118.
also inside colyfOn the web and on mobile, Lamp answers from what you’re allowed to see and proposes changes for you to confirm.Lamp in Colyf
how it replies
Short replies. Real work behind them.
Lamp answers in a sentence or two. It doesn’t list its features, cheer you on or ask whether you’d like it to check. It checks, then tells you.
- Greets you by name, then gets out of the way.
- Asks one short question when something essential is missing.
- Never reports a change it hasn’t read back.
Lampbot
move the launch ticket to in progress
Done. CLM-42 is in progress.
Example conversation. Names and tickets are made up; the steps are the ones Lamp runs.
- request
- “move the launch ticket to in progress”
- route
- Main (CLM)default project
- find
- CLM-42 · Launch checklist
- read
- status: To Do
- change
- To Do → In Progress
- read back
- status: In Progress✓
- reply
- “Done. CLM-42 is in progress.”
jira, carefully
It reads before it writes.
Lamp checks a ticket before changing it and reads it back afterwards. Before creating one, it looks for duplicates. If a write times out, it checks whether the change landed before trying again.
- Search and read tickets
- Create tickets, after checking for duplicates
- Edit summaries and descriptions
- Comment
- Move a ticket’s status
- Assign by name or email
- Look people up in your Jira directory
Account IDs stay inside the integration; Lamp uses names and whatever emails Jira shares. Work goes to your default project unless the request names another one or matches a routing rule you set.
memory
Memory that belongs to one person.
Each person gets a private profile, a page of notes and a journal. Lamp saves the useful facts as it works and can look back through earlier conversations. It won’t claim to remember what isn’t written down.
- Private to each person. Other users can’t read it through Lamp.
- Plain Markdown you could read yourself, up to 256 KiB per file.
- Older journal entries move to an archive instead of being deleted.
Who you are and how you like to work.
# Ana Ruiz - Prefers short updates - Works on: mobile app, release planning - Time zone: CET - Calls tickets “cards”
Users/your-id/PROFILE.md · example
Stable facts and decisions Lamp saves while it works.
- Release moved to Friday after the QA review - Weekly report goes to the #ops channel - Default board for design work: DES - Don’t assign tickets to the contractors
Users/your-id/MEMORY.md · example
The conversation itself, journaled automatically. Older entries move to an archive.
## Oct 14 **Ana:** what’s blocking the release? **Lamp:** QA sign-off on MOB-131. ## Oct 15 **Ana:** send me today’s standup notes **Lamp:** Here you go. standup-notes.md
Users/your-id/JOURNAL.md · example
files
Files in, files out.
Send Lamp a text file and it reads it. Ask for one and it writes the file in your private folder, then sends it back into the chat.
| in | Markdown, text, CSV, JSON and code files | up to 1 MB each |
|---|---|---|
| out | Files Lamp writes for you | sent back as chat attachments |
| space | A private folder per person | 8 MiB · 256 files |
| not yet | PDF, Word and images | planned |
boundaries
Lit where you’re working. Dark everywhere else.
Lamp gets a short list of tools and nothing else. Every call is checked against that list three times: when tools are offered to the model, when a call is dispatched and when it runs. Anything outside it is refused before it starts.
try to make lamp misbehave · simulated in your browser
Pick one. The lamp will tell you.
Can use
Your folder and notes
- lamp_read
- lamp_write
- lamp_list
- lamp_memory_read
- lamp_memory_write
- lamp_recall
Jira
- jira_route
- jira_search
- jira_get
- jira_create
- jira_update
- jira_comment
- jira_transitions
- jira_transition
- jira_assign
- jira_members
- jira_find_user
Switched off
- Shell and terminal
- SSH
- Web browsing
- Running code
- Handing work to other agents
- Scheduled automations
- Changing its own settings
- Shared memory between people
Refused paths
- Absolute paths
- ../ traversal
- Hidden files
- Device names like CON or NUL
- Symlinks, junctions and hard links
These limits are enforced by the application, not by an operating-system jail. We’d rather tell you that than oversell it. Jira actions use the connected account’s permissions, so connect an account with the access you intend.
under the hood
If one model fails, another answers.
Lamp runs on a primary model with an automatic fallback to a second provider. It switches when the provider fails, not as a race on every message, so you get one reply either way.
Primary provider answering.
- Encrypted snapshots
- every 60 seconds, the last five kept
- One worker at a time
- so two copies never answer the same chat
- Encrypted backups
- you can download and restore
- Model names
- never shown in chat; the provider list is available on request
not yet · october 2026
What Lamp doesn’t do yet.
- Read PDFs, Word documents or images.Text, Markdown, CSV, JSON and code files work today.
- Run inside an operating-system sandbox.Today’s limits are enforced by the application.
- Let you sign up on your own.Access is by request for now.
- Remember everything ever said.It recalls saved notes and past conversations, and says so when it can’t find something.
for teams, agencies and schools
Plain answers before you ask.
If something isn’t settled for your team, we’ll say so and put the answer in writing.
Ask us anythingWhich chat apps does Lamp work in?
Telegram, WhatsApp and Slack. Inside Colyf, Lamp also works on the web and on mobile, with its own separate setup.
Who can see my notes?
Your profile, notes and journal belong to you, and other users can’t read them through Lamp. The people who run your Lamp deployment manage the server it runs on, so ask us how that’s set up for your team.
Does Lamp train on our conversations?
Lamp doesn’t train models. Requests are processed by model providers under their API terms. Ask us for the provider list and those terms.
Where is our data stored, and for how long?
Ask us. We’ll put the hosting location, retention and deletion process in writing for your team before you start.
Can Lamp change things without asking?
It acts on clear requests and tells you what it did. It reads a ticket before changing it and reads it back after. When something essential is missing, it asks one short question first.
What can Lamp do in our Jira?
What the connected account can do, limited to the actions Lamp has: search, read, create, edit, comment, assign and move status. There is no delete. Connect an account with the access you intend.
Is this the same assistant that’s inside Colyf?
They’re separate integrations. In Colyf, Lamp works inside the platform on web and mobile and proposes changes you confirm. The assistant in Telegram, WhatsApp and Slack has its own identity, memory and tools.
How do we get access?
Email us with your team size, chat app and Jira setup. We’ll reply with the next steps.
Make light work of it.
Lamp is available by request for teams, agencies and schools.